Human pose estimation is critical in computer vision, with applications in augmented reality, animation, gaming, and robotics. This paper presents a comprehensive approach using deep neural networks, leveraging the BlazePose model through the MediaPipe library. Our method processes depth images to obtain 3D landmarks and background segmentation for high-fidelity body pose tracking. We explore neural network architectures, including convolutional neural networks (CNNs) and recurrent neural networks (RNNs), to capture spatial dependencies and handle multi-scale features. The dataset is divided into training, validation, and test sets, and the model is trained using stochastic gradient descent (SGD). We evaluate performance using metrics such as Percentage of Correct Keypoints (PCK) and Percentage of Detected Joints (PDJ), with visualizations of predicted keypoints on input images. Our approach is integrated into applications like activity recognition, gesture recognition, and human–computer interaction, demonstrating the efficacy of deep neural networks in precise and robust human pose estimation for real-time interactive systems.